732471 - low fps (4 - 17) on http://soulwire.github.com/Coffee-Physics/

Status: RESOLVED WORKSFORME

(Core :: JavaScript Engine, defect)

Description

[Dietrich Ayala (:dietrich)]

Getting very poor performance on Nightly for these demos.

Comment 1

[Thomas Ahlblom]

Reproduced: (slow, and I also miss the colored bubbles)

15-16 fps
Mozilla/5.0 (X11; Linux x86_64; rv:10.0.2) Gecko/20100101 Firefox/10.0.2

15-16 fps
Mozilla/5.0 (X11; Linux x86_64; rv:11.0) Gecko/20100101 Firefox/11.0

14-38 fps (initial spike, then stabilized at 15 fps)
Mozilla/5.0 (X11; Linux x86_64; rv:12.0a2) Gecko/20120302 Firefox/12.0a2

15-17 fps
Mozilla/5.0 (X11; Linux x86_64; rv:13.0) Gecko/20120302 Firefox/13.0a1  



Works for me:

67-81 fps
Opera/9.80 (X11; Linux x86_64; U; en) Presto/2.10.229 Version/11.61

61-62 fps
Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/535.11 (KHTML, like Gecko) Chrome/17.0.963.56 Safari/535.11

Comment 2

[Andrew McCreight [:mccr8]]

I'm getting around 50fps on my Mac.

Comment 3

[Dietrich Ayala (:dietrich)]

I'm getting 2-16fps on Mac w/ Nightly.

Comment 4

[Boris Zbarsky [:bzbarsky]]

Interesting.  I get pretty good fps (50+) on Mac in a clean profile with just the one tab open, but crappy fps in my normal browsing profile with lots of other tabs....

Comment 5

[Boris Zbarsky [:bzbarsky]]

The time usage I _do_ see is almost all JS stuff.  mjit code, slow calls for array_push, realloc/malloc, etc.  That's about 85% of the time.

Comment 6

[David Mandelin [:dmandelin]]

I don't seem to repro this on Windows, but I do on Mac. So it could be an x64 jitcode issue. The strange thing is that it seems to generally run well, but experience long pauses that seem to correlate with GCs. But the GC pauses are short, so if it is related to GC, it's probably recompilation.

Taking off tracking14 because this is yet another perf bug. It'll go on k9o, though. Bill, if you want to take a quick look with your visualizer, you might turn up something interesting.

Comment 7

[Nick Cameron [:nrc]]

I got a comment on my blog about this (http://www.blogger.com/comment.g?blogID=36017112&postID=5248071291525468244), apparently the demo was slow with HWA on, but fast with it off on Windows, we've had this problem before due to gradient caching, but there don't seem to be any gradients in this demo.

I couldn't reproduce this behaviour, but the demo runs like an absolute dog in an optimised build (Nightly, v19) on my fairly well-speced machine - very low fps and enormous (~5sec) GC pauses. I get the same behaviour with and without HWA and with different combinations of D2D and software (Cairo) canvas (I mainly tested the canvas version of the demo). The demo runs pretty nicely on my Beta (17) build, same config and using D2D (Azure) canvas, so I don't think this is an Azure/Thebes canvas regression.

Re bz's comment above, profile for testing on Beta had a lot of tabs open, ~60. Testing on Nightly was with 4, and two of those were about:*

Comment 8

[Mayank Bansal]

Nick , i filed https://bugzilla.mozilla.org/show_bug.cgi?id=803957  regarding the issue i reported on your blog.

Comment 9

[Jan de Mooij [:jandem]]

I get a stable 60 fps here on OS X and Windows with the latest nightly.

With Ion disabled, it's about 15 fps and with Baseline disabled it's 1-2 fps, so we are definitely running a lot of JS but it seems to be Ion-compiled just fine.

Please reopen or file a new bug if you can still reproduce this.